locking/atomic/x86: Introduce arch_atomic64_try_cmpxchg() to x86_32
Introduce arch_atomic64_try_cmpxchg() for 32-bit targets to use optimized target specific implementation instead of a generic one. This implementation eliminates dual-word compare after cmpxchg8b instruction and improves generated asm code from: 2273: f0 0f c7 0f lock cmpxchg8b (%edi) 2277: 8b 74 24 2c mov 0x2c(%esp),%esi 227b: 89 d3 mov %edx,%ebx 227d: 89 c2 mov %eax,%edx 227f: 89 5c 24 10 mov %ebx,0x10(%esp) 2283: 8b 7c 24 30 mov 0x30(%esp),%edi 2287: 89 44 24 1c mov %eax,0x1c(%esp) 228b: 31 f2 xor %esi,%edx 228d: 89 d0 mov %edx,%eax 228f: 89 da mov %ebx,%edx 2291: 31 fa xor %edi,%edx 2293: 09 d0 or %edx,%eax 2295: 0f 85 a5 00 00 00 jne 2340 <...> to: 2270: f0 0f c7 0f lock cmpxchg8b (%edi) 2274: 0f 85 a6 00 00 00 jne 2320 <...> Signed-off-by: Uros Bizjak <ubizjak@gmail.com> Signed-off-by: Ingo Molnar <mingo@kernel.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Link: https://lore.kernel.org/r/20240410062957.322614-1-ubizjak@gmail.com
Showing
Please register or sign in to comment